Affinity-Aware Sharding for Delayed Tensor Parallelism
Read the original on arXiv Machine Learning →The Flow has not summarised this story yet — read it at arXiv Machine Learning.
The Flow has not summarised this story yet — read it at arXiv Machine Learning.
arXiv:2606. 09377v1 Announce Type: cross Abstract: Formal neural network verification -- proving that a network satisfies safety properties for \emph{all} inputs in a specified domain -- is bounded in practice by GPU memory: standard implementations of bound-propagation algorithms (IBP, CROWN, $\alpha$-CROWN) require weight and relaxation-coefficient matrices to reside entirely on one accelerator.
Formal neural network verification -- proving that a network satisfies safety properties for \emph{all} inputs in a specified domain -- is bounded in practice by GPU memory: standard implementations of bound-propagation algorithms (IBP, CROWN, $α$-CROWN) require weight and relaxation-coefficient matrices to reside entirely on one accelerator. We adapt two parallelism techniques originally developed for large-scale model training to the \texttt{auto\_LiRPA}\,/\,$α,β$-CROWN verification framework.
arXiv:2609.37899v1 Announce Type: new Abstract: Zero-order optimization (ZO) trains without backpropagation, making it relevant to forward-only hardware and non-differentiable loss, but its gradient...
BellowsKV introduces a training‑free approach that maps the hidden states of multiple adjacent layers in a large language model to a single shared latent representation per token, allowing each layer to retrieve its own keys and values via a layer‑specific projection. The method retains the latent at full dimension and offers two compression strategies: BellowsKV‑Quant, which stores the latent at low precision, and BellowsKV‑Evict, which makes a shared eviction decision per block. Experiments on Llama‑3.1‑8B, Mistral‑7B, and Qwen2.5‑7B show that BellowsKV‑Quant preserves about 99% of full‑cache accuracy at 3.6× compression and 97‑98% at 13‑14×, outperforming existing cross‑layer methods that drop to at most 84% accuracy at 16× compression.
arXiv:2605. 15250v3 Announce Type: replace-cross Abstract: Multi-head Latent Attention (MLA), the attention used in DeepSeek-V2/V3, jointly compresses keys and values into a low-rank latent and matches the H100 roofline almost perfectly.
The paper introduces Decode Token Parallelism (TKNP), a new decoding architecture that separates model weight placement from attention computation. TKNP distributes a single model copy across nodes in balanced layer blocks, allowing each node to compute attention for its own requests using local KV caches while performing dense operations within a single NVLink domain. Implemented in vLLM, TKNP achieves up to 1.7× higher throughput than decode context parallelism and up to 2.06× over data parallelism on large models such as Llama 3.1 405B and Nemotron‑4 340B across 64 H100 GPUs.